SafeBite AI: An Offline Ingredient Safety Companion Built for Ananya

A developer built SafeBite AI, an offline ingredient-safety scanner that runs Google's Gemma 2 open-weight model locally via Ollama to give a SAFE, WARNING, or UNSAFE verdict against a user's allergy and dietary profile. The Python 3.10 tool was created for a friend with severe peanut allergies and lactose intolerance so packaged-food labels can be checked without internet access, subscriptions, or sending health data to remote servers.

This is a submission for the Hacktoberfest Weekend Challenge: Build for a Friend https://dev.to/challenges/hacktoberfest-weekend-2026-10-01 I built SafeBite AI for my close friend and roommate, Ananya. Ananya lives with severe peanut allergies and acute lactose intolerance, and she follows a strict Pure Vegetarian Eggless diet. Shopping for packaged snacks in local Indian grocery stores is a daily gamble for her. Ingredient lists are printed in microscopic font on shiny plastic packaging, hiding severe allergens behind complex terms milk solids, peanut derivatives or vague additive numbers INS 322, E120 . She routinely spends minutes staring at labels under her phone light, terrified of an accidental allergic reaction or consuming non-veg additives. Commercial scanner apps require constant internet access which fails inside grocery basements , charge subscriptions, and sell health data. SafeBite AI is a lightweight, local AI assistant that scans ingredient lists and instantly gives an unambiguous SAFE, WARNING, or UNSAFE verdict tailored specifically to Ananya's profile. ================================================== Input: Refined Wheat Flour Maida , Sugar, Palm Oil, Peanut Oil, Milk Solids, INS 322. VERDICT: 🛑 UNSAFE FOR ANANYA ANALYSIS & REASONS: RECOMMENDATION: Do not consume. Pick a dairy-free, nut-free alternative. https://github.com/yasmeenvrk23-beep/Safebite/blob/main/README.md https://github.com/yasmeenvrk23-beep/Safebite/blob/main/README.md •Open-Weight AI Engine: Powered by Google's Gemma 2 open-weight model running locally via the Ollama framework. •Language & Environment: Built with Python 3.10 and executed entirely locally. •Custom System Prompting: Configured with a dedicated rule-engine prompt that parses messy, comma-separated food packaging text against severe allergy and dietary constraint flags. Using open-weight AI models and open-source tooling was non-negotiable for this project: •Supermarket Basement Reliability: Grocery store basements in India frequently suffer from zero cellular signal. Running an open-weight model locally Gemma means Ananya can check safety offline with zero network latency or connectivity drops. •Absolute Health Privacy: Ananya's medical conditions, allergy severity, and personal dietary restrictions stay entirely on her local device rather than being logged on remote servers. •Zero Maintenance Costs: As students, using open weights locally removes expensive cloud API token fees, enabling a completely free tool for daily life. Built using direct terminal/Google Colab local execution sessions with Gemma 2 •Best Use of Gemma $200 : Uses Google's Gemma open-weight model to perform fast, private, and offline label scanning. •Best Use of GitHub $100 : Built, version-controlled, and publicly documented open-source on GitHub.